“Would a 3% price increase for every customer bring in more than 5% for the 80% who spend least?”
A decision, not a lookup. The kind usually settled on gut feel in a board meeting, because nobody has time to pull the evidence together.
The questions that matter usually need more than one system. The AI layer reads your ERP, CRM and other connected systems together, shows where each answer came from, and keeps a record of every request. It only ever reads.
Your ERP knows what customers ordered and what it cost. Your CRM knows who sales spoke to. The question on the table usually needs both, and pulling them together by hand takes longer than the meeting allows.
Dashboards help with questions someone set up in advance. The ones that matter in a board meeting are usually new each time.
You type the question the way you would ask a colleague who knows the whole business.
“Would a 3% price increase for every customer bring in more than 5% for the 80% who spend least?”
A decision, not a lookup. The kind usually settled on gut feel in a board meeting, because nobody has time to pull the evidence together.
“What objections have come up most in customer conversations over the last 90 days?”
The answer is in conversations, not in a number field. It can be answered where those conversations are held in a connected system.
“Which of our larger accounts have gone quiet with sales this quarter?”
Order history is in the ERP and sales activity is in the CRM. Neither can answer it alone.
“How many times has customer data been put into AI by staff this month?”
Because every request goes through one place, the business can see its own use of AI.
Staff may already be putting business data into AI tools, and most of those tools keep no record the business can see. The AI layer does, for every request made through it.
So when a customer, an auditor or a supplier questionnaire asks what your AI touched, you have the record.
| Time | Who | Request | Systems used |
|---|---|---|---|
| 09:15 | Sales | Summarised a key account’s order history | ERP, CRM |
| 11:40 | Finance | Listed invoices more than 60 days overdue | ERP |
| 14:02 | Operations | Compared supplier lead times over the last 6 months | ERP, spreadsheet |
| 16:27 | Managing director | Checked whether the data supports a price change for smaller accounts | ERP, CRM |
An AI that can change your ERP can put wrong data into the system the business runs on. The AI layer only reads. Where a bespoke module needs to write back, every field it can touch is named in the specification you approve, and most builds need no write access at all.
It reads. Where a bespoke module writes back, every field it can touch is named in the specification you approve.
Nothing runs unless the specification says it runs.
You hear “the data cannot answer that”, and why, rather than a confident guess.
Each answer shows which systems and records it came from, so it can be checked.
Your ERP, CRM and spreadsheets stay as they are. Sysgraft works on top of them.
A UK wholesaler, about £10m turnover. Questions answered across the company’s systems, with every use of AI on the record.
Live in 8 days. Bought by the managing director.
Order matters. The business problem comes first, then the system, then the AI. The layer is only as useful as the systems joined underneath it, which is why it is usually built after Curavest has found what needs fixing.
Most Sysgraft work begins with Curavest’s Operational Improvement Workshop: two hours on one problem that keeps coming back, on site or on Teams. No obligation, and no records needed. If the answer turns out to need software, that is a separate decision, for later.